Exploring How Large Language Models Understand Sexual Communication in Hypothetical Sexual Situations

Large language models (LLMs), or models which generate human like responses to people’s queries, are growing in popularity and may be an additional tool to access information related to sexual communication—including sexual consent and refusals. However, it remains unclear whether LLMs can identify sexual communication in hypothetical situations and how this influences their understanding of whether an encounter could be consensual, despite the importance of this capability as these tools become increasingly integrated into the sexual health field. Our study provided four different LLMs a series of seven sexual scenarios where sexual communication and context was manipulated. The models were asked to evaluate if an experience was consensual based on communication cues and context and explain why. We then compared the responses from the LLMs with eight experts in gender-based violence. When scenarios depicted clear and verbal consent or refusal communication, LLMs were consistently able to identify if the experience was consensual and matched with the experts. Experts and LLMs disagreed on when intoxication presented challenges for sexual communication and understandings of consensual sexual experiences. Our results suggest LLMs may struggle in their interpretation of complex, nuanced scenarios which require improvement to align more closely with expert judgment. Future research should focus on enhancing LLM training to better handle the subtleties of human behavior and context during instances of sexual communication.